Enhancing Short-Term Wind Speed Prediction Based on Deep Learning With Ensemble Learning Model for Small Wind Turbine Applications

Wind is unstable and unpredictable, and power generation is not constant. Wind speed prediction reduces these disadvantages, and it is essential to measure accurate wind speed predictions to install and stabilize wind power generation systems. This research focused on predicting wind speed data collected from various locations in India, such as Amaravati, Bangalore, Kanyakumari, and Kochi, at heights of 10 m and 50 m, with a sampling period of 1 h for small wind turbine (SWT) applications. This study discusses various deep learning (DL) algorithms for enhancing wind speed forecasting accuracy. The performance of deep learning algorithms is evaluated using multiple metrics, namely, mean square error, normalized mean square error, root mean square error, normalized root means square error, relative root mean square error, mean absolute percentile error, symmetric mean absolute percentage error, and coefficient of determination. For the short-term wind speed prediction, the ensemble learning model approach gave the best results in all sites and among all models applied. The results of the wind speed prediction algorithms show that the Kanyakumari datasets have improved accuracy compared to other locations. At this location, the R2 value is about 0.9942 at a height of 10 m and 0.9955 at 50 m. Further, this dataset is segregated into seasons and months. During the summer seasons, short-term wind speed predictions are more accurate, with R2 values of 0.9885 at 10 m and 0.9904 at 50 m, and November stands out for its highest efficiency in monthly wind speed forecasts, with R2 values reaching 0.9910 at 10 m and 0.9919 at 50 m.